Infrastructure in northern regions is increasingly threatened by climate change, mainly due to permafrost thaw. Prediction of permafrost stability is essential for assessing the long-term stability of such infrastructure. A key aspect of geotechnical problems subject to climate change is addressing the surface energy balance (SEB). In this study, we evaluated three methodologies for applying surface boundary conditions in longterm thermal geotechnical analyses, including SEB heat flux, n-factors, and machine learning (ML) models by using ERA5-Land climate reanalysis data until 2100. We aimed to determine the most effective approach for accurately predicting ground surface temperatures for climate-resilient design of northern infrastructure. The evaluation results indicated that the ML-based approach outperformed both the SEB heat flux and n-factors methods, demonstrating significantly lower prediction errors. The feasibility of long-term thermal analysis of geotechnical problems using ML-predicted ground surface temperatures was then demonstrated through a permafrost case study in the community of Salluit in northern Canada, for which the thickness of the active layer and talik were calculated under moderate and extreme climate scenarios by the end of the 21st century. Finally, we discussed the application and limitations of surface boundary condition methodologies, such as the limited applicability of the n-factors in long-term analysis and the sensitivity of the SEB heat flux to inputs and thermal imbalance. The findings highlight the importance of selecting suitable boundary condition methodologies in enhancing the reliability of thermal geotechnical analyses in cold regions.

Permafrost on the Tibetan Plateau (TP) is controlled by high-elevation and the complex hydrothermal processes and energy balance on the ground surface. To successfully model or map permafrost distribution, it is necessary to parameterize near-surface air or land-surface temperatures (Ta or LST) to ground surface temperature (GST) at local-, meso-, or macro-scale. Here, a long-term experimental observation (November 2010 to December 2018) was conducted for understanding the differences between Ta and GST at a plot with 26 sites at Chalaping to the south of the Sisters Lakes in the Source Area of the Yellow River, northeastern TP. Results show that GST varies considerably within an area of about 3.5 km2 under the combined thermal influences of surface vegetation, soil moisture conditions, and microtopography. Mean annual GST (MAGST) ranged from -0.55 to -3.02 degrees C, with an average of -1.35 +/- 0.63 degrees C. The surface offset varied from 1.01 to 3.90 degrees C, with an average of 2.72 +/- 0.70 degrees C. The difference between monthly Ta and monthly GST decreased from 4.64 +/- 2.09 degrees C in January to 1.09 +/- 1.34 degrees C in July and then gradually increased to 5.61 +/- 2.53 degrees C in November. The active layer thickness (ALT) calculated with the ground-surface thawing index ranged from 0.85 to 1.95 m, with an average of 1.51 +/- 0.33 m. Annual freezing N-factors and annual thawing N-factors were averaged at 0.58 +/- 0.12 and 1.31 +/- 0.28, respectively. Although weakly, hourly and daily GST values are positively correlated to NDVI, while ALT negatively correlated with NDVI. This study demonstrates the complex thermal regimes on the ground surface, even within a small area despite the relatively consistent topography. It will likely facilitate the parameterization of the upper thermal boundary of permafrost modeling or mapping on the TP where the landscapes are characterized by extensive presence of dwarf alpine meadow and alpine steppe, further contributing to the study in ecosystem feedbacks to the regional climate change.